BROWNIAN DYNAMICS SIMULATIONS

OF POLYMERS AND SOFT MATTER

The Brownian dynamics (BD) simulation technique is a mesoscopic method in which explicit
solvent molecules are replaced instead by a stochastic force. The technique takes advantage
of the fact that there is a large separation in time scales between the rapid motion of solvent
molecules and the more sluggish motion of polymers or colloids. The ability to coarse-grain
out these fast modes of the solvent allows one to simulate much larger time scales than
in a molecular dynamics simulation. At the core of a Brownian dynamics simulation is a
stochastic differential equation which is integrated forward in time to create trajectories of
molecules. Time enters naturally into the scheme allowing for the study of the temporal
evolution and dynamics of complex fluids (e.g. polymers, large proteins, DNA molecules and
colloidal solutions). Hydrodynamic and body forces, such as magnetic or electric fields, can
be added in a straightforward way. Brownian dynamics simulations are particularly well
suited to studying the structure and rheology of complex fluids in hydrodynamic flows and
other nonequilibrium situations.

Basic Brownian Dynamics

The technique of Brownian dynamics is used to simulate the dynamics of particles that
undergo Brownian motion. Because of the small mass of these particles, it is common to
neglect inertia. Using Newtons Second Law for particle i, Ftot i = mi ai , the neglect of
inertia means that the total force is always approximately zero. The total force on a particle
is composed of a drag force Fdi from the particle moving through the viscous solvent, a
Brownian force FBi due to random collisions of the solvent with the particle, and all non-
hydrodynamic forces Fnhi
Ftot d B nh
i = Fi + Fi + Fi 0 . (1)
This total non-hydrodynamic force includes any external body forces, any spring forces, and
any excluded volume interactions. In creeping flow and neglecting hydrodynamic interactions
(free-draining), the drag force is taken as Stokes drag on a sphere
 
d dri
Fi = u (ri ) (2)
dt

where is the drag coefficient and u (ri ) is the unperturbed velocity of the solvent evaluated
at the position of the particle. The differential equation governing the motion of the particle
then becomes
dri 1 nh
Fi ({rj }) + FB


= u (ri ) + i (t) (3)
dt

1
and is commonly called a Langevin equation. Note that the non-hydrodynamic force depends
on the set of all particle positions {rj }. This is a stochastic differential equation because the
Brownian force is taken from a random distribution. In order for the dynamics to satisfy
the fluctuation-dissipation theorem, the expectation values of the Brownian force are

hFB
i (t)i = 0 (4)

hFB B
i (t)Fj (t )i = 2kB T ij (t t ) (5)
where kB is Boltzmanns constant, T is the absolute temperature, ij is the Kronecker delta,
(t t ) is the Dirac delta function, and is the unit second-order tensor. Equations (3),
(4), and (5) are equivalent to the Fokker-Plank equation description, which is a diffusion

equation for the phase space probability density (Ottinger, 1996).
Having developed the governing stochastic differential equation, one performs a BD simu-
lation by integrating this equation forward in time. The stochastic nature means that one
must produce many independent trajectories that are averaged together, producing the time-
evolution of an ensemble-averaged property. The repetition of many independent trajectories
is a time-consuming but necessary part to follow the time-evolution of a property. However,
to calculate a steady state property, one uses the ergodic hypothesis to time-average a single
trajectory.

Hydrodynamic Interactions

As a particle moves along its trajectory, it exerts a force on the solvent which changes
the velocity field from its undisturbed value. The disturbance velocity changes the viscous
drag force exerted on the other particles. This interaction between particles mediated by the
solvent is called hydrodynamic interaction (HI). The hydrodynamic interactions are included
in Brownian dynamics through the use of an interaction tensor ij included as part of the

diffusion tensor Dij (Ottinger, 1996). The two tensors are related by

kB T
Dij (ri , rj ) = (ij + ij (ri , rj )) (6)

where ii = 0. The stochastic differential equation including HI then becomes

N N
dri 1 X nh
X
= u (ri ) + Dij (ri , rj ) Fj ({rk }) + 2 Bij ({rk }) nj (t) (7)
dt kB T j=1 j=1

where nj are random vectors with expectation values

hnj (t)i = 0 (8)

2
 hni (t)nj (t )i = ij (t t ) (9)
and the weighting factors Bij must be calculated from the diffusion tensor in order to satisfy
the fluctuation-dissipation theorem
N
X
Dij (ri , rj ) = Bip ({rk }) BTjp ({rk }) . (10)
p=1

This can be inverted to calculate Bij by Cholosky decomposition. However, a more efficient
method has been developed by Fixman (1986) and implemented in BD simulations.
The interaction tensor in an unbounded solvent is taken as the Rotne-Prager-Yamakawa
tensor, which is a regularized version of the Oseen-Burgers tensor. The Oseen-Burgers
tensor represents the disturbance due to a point-force in creeping flow. However, it results in
a non-positive-definite diffusion tensor if particle separations are comparable to the particle
radius. The Rotne-Prager-Yamakawa tensor modifies the small separation disturbance such
that the diffusion tensor is always positive-definite.

Polymer Models used in Brownian Dynamics

The choice of polymer model is intrinsically a modeling decision which depends upon the
real polymer one wants to model and the level of fine-scale molecular detail one needs to
retain or can computationally afford to simulate. Polymers can be broadly separated into
flexible and semiflexible chains. The flexibility of a chain is determined by the ratio L/lp
where lp is the persistence length and L the contour length of the chain. Flexible chains
have L/lp 1 and semiflexible chains have L/lp 1.
The most common coarse-grained models for flexible polymers are the freely-jointed bead-
rod and bead-spring chains. The polymer is modeled as a series of beads connected by either
rods or springs, as shown in Figure 1. The frictional drag on the chain is distributed at bead
centers. The term freely-jointed implies that there is no energetic penalty to rotating a
spring or rod about a bead center. The spirit of these mesoscopic models is to coarse-grain
out molecular details smaller than the finest length scale in the given model (rod or spring).
We consider first the flexible bead-rod chain. Physically, the rod in a bead-rod chain cor-
responds to a Kuhn length lk (twice the value of the persistence length). Mathematically,
the rods act as a constraint on the system which ensures that adjacent beads in the chain
are maintained at a constant separation at all times. How one achieves this constraint is
important as there is a subtle difference between a completely rigid constraint and the ap-
proximation of that constraint using a very (infinitely) stiff potential (Morse, 2004). For
example, the equilibrium distribution of a bead-rod chain using stiff constraints yields a

3
Figure 1: Schematic of the canonical polymer model used in Brownian dynamics simulations:
a bead-spring chain. The continuous curve is the real polymer that the bead-spring chain
is meant to represent.

random walk configuration while imposing rigid constraints gives rise to correlations in the
rod directions, even in the absence of bending forces. Physically, one would argue that the
random walk configuration is more realistic. However, rigid constraints are attractive from
a computational standpoint as they freeze out the usually uninteresting rapid bond fluctu-
ations. In practice, one chooses to simulate the stiff system, but does so by imposing rigid
constraints (which introduces a new tension force Ftens ) and adding a corrective pseudo-
potential metric force Fmet which makes the system equivalent to imposing an infinitely stiff
potential. A recent detailed review of constrained Brownian motion and the implementation
of bead-rod BD algorithms can be found in Morse (2004). With current computers typically
one can only simulate chains with up to 100 200 Kuhn steps which corresponds to a low
molecular weight polystyrene polymer ( 75, 000 150, 000 Daltons).
Large flexible polymers are more commonly modeled as a series of Ns springs connected by
beads. Each spring models a portion of the full chain and has a contour length Ls = L/Ns .
The spring represents the entropic restoring force associated with stretching a subsection
of the chain. The entropic force can be calculated starting from a fine-scale micromechan-
ical model (e.g. freely-jointed bead-rod chain) using equilibrium statistical mechanics and
calculating the extension of a chain when subject to a constant force. The force-extension
response of a freely-joined chain is exactly described by the inverse Langevin function which
is closely approximated by the more convenient FENE force law

3kB T r/Ls
F F EN E (r) = (11)
lk [1 (r/Ls )2 ]

where r is the spring extension and Ls is the spring length when fully stretched (Bird et al.,

4
1987). A slightly more accurate approximation to the inverse Langevin function can be
obtained using a Pade approximation (see for example Hsieh et al. (2003)). Most flexible
synthetic polymers (polystyrene, poly(ethylene oxide) etc.) and single stranded DNA have
significant bond rotations and should be modeled using the FENE force law. However, many
biopolymers resist local bond torsion (e.g. duplex DNA) and are more accurately described
by a wormlike chain model which for Ls /lp 1 is well approximated by the Marko-Siggia
spring law (Marko and Siggia, 1995)
"  2 #
k B T 1 r 1 r
F wlc (r) = 1 + . (12)
lp 4 Ls 4 Ls

It is important to note some limitations and assumptions when using common spring force
laws such as Equations (11) and (12). These relations are derived assuming a large number
of persistence or Kuhn lengths are contained in the spring. More refined calculations show
however that corrections to the spring laws must be made when the spring has a contour
length less than 20lk . Furthermore, if the underlying micromechanical model is freely-
jointed at each node connecting the springs (e.g. freely-jointed chains), then spring force laws
can be derived for springs containing an arbitrarily small number of Kuhn lengths which will
exactly reproduce the force extension response of the micromechanical model. The use of
equilibrium statistical mechanics to derive spring forces also raises a subtle point when using
a bead-spring model in flow or other nonequilibrium situations. In such cases it is implicitly
assumed that the springs are deformed slowly enough such that the micromechanical model
describing the spring would be able to fully sample its configuration space; in a sense
achieving a local equilibration in phase space. Lastly, it is important to note that parameters
in the spring force are directly related to the physical polymer system being modeled (e.g. a
certain subsection of a polymer chain of length Ls with a given persistence length lp ). An
extensive discussion of these and other issues on the coarse-graining polymers into bead-
spring chains is given in Underhill and Doyle (2004).
The model for a semiflexible polymer is based on the premise that the polymer can be
described as a homogeneous, isotropic elastic filament. The energy associated with bending
a continuous filament described by a curve r(s) is U bend = /2 ds | 2 r/s2 |2 where s is a
R

contour distance along the chain. The bending rigidity is related to the persistence length
by lp = /kB T . The chain is typically course-grained to a series of beads connected by rods
which have a bending energy given by
N 1
bend X
U = uj uj1 (13)
a j=2

5
where a is the length of the connecting rod and uj is the unit vector directed from bead j to
j + 1. The bending force is Fbend
i = U bend /ri . This description of a semiflexible polymer
is commonly referred to as the wormlike chain or Kratky-Porod model (Yamakawa, 1997).
Systems with constraints (rods between beads) which ensure local inextensibility of a chain
are computationally expensive and are sometimes replaced by stiff Fraenkel springs of the
form Fsi = H(|ri+1 ri | a)ui where H is the spring constant and a is the equilibrium length
of the spring. Pasquali and coworkers have shown that rigid constraints and stiff Fraenkel
springs give similar results for the collapse of DNA in poor solvent (Montesi et al., 2004).
They found typical time savings in using stiff springs over rigid constraints are on the order
of 10 50 fold. Most researchers use stiff springs only when simulating a polymer which
is at or near equilibrium. Detailed comparisons between Fraenkel springs and constraints
have not yet been performed for polymers in flow or under large tensions. In general, it is
recommended to use rigid rods (with the corrective metric force discussed previously) unless
one is certain stiff springs introduce no computational artifacts.

Numerical Algorithms

We have developed the stochastic differential equation and discussed the types of models
and corresponding forces involved. To solve for the trajectory, one must integrate forward
in time. A simple explicit Euler time integration scheme is widely used
 
dri
ri (t + t) ri (t) + (t) t . (14)
dt

In this discrete version, the random vectors have an expectation value

hni (t)nj (t )i = ij tt /t . (15)

If the forces are steep, this would require very small time steps. This is particularly important
for bead-spring chains with finitely extensible springs such as the FENE or Marko-Siggia force
law. The above algorithm allows for the possibility of a spring being stretched beyond its
fully-extended length. A semi-implicit predictor corrector method including HI has been
developed that prevents this overstretching of springs (Hsieh et al., 2003).
The realization of a trajectory requires the sampling of random vectors with the appropriate
expectation values. The calculation of these random numbers is a significant computa-
tionally cost of BD simulations. Because of this cost, it is better to use uniform random
numbers than Gaussian distributed. These random number generators should be used with

care. Ottinger (1996) reviews important aspects of random number generators.

6
The use of random variables in the simulation and a finite number of trajectories in the
ensemble means that there is intrinsic statistical noise to the method. The size of this
1/2
error is proportional to NT where NT is the number of independent trajectories. An
important technique for the reduction of this error by reducing the proportionality factor
(not by increasing NT ) is called variance reduction. The exact way of performing variance
reduction depends on the system of study. Two important types of variance reduction are

importance sampling and subtracting off a control variable (Ottinger, 1996).
Another issue to be noted for the implementation of BD algorithms concerns the neglect of
mass, which is a singular limit. This limit is discussed in detail by Grassia et al. (1995). We
note here that one consequence of this singular limit is a drift that results if the diffusivity
depends on position. To correct for this drift, a term with the gradient of the diffusivity must
be added to the algorithm. However, if a higher order scheme is used such as a midpoint
method or predictor corrector method, the extra term should not be added.

Computing Stress

The rheology of most complex fluids is not described well by simple constitutive relations.
Central to the study of rheology then is the calculation of the bulk stress tensor in a
simulation. The bulk stress of the mixture is a linear combination of the solvent contribution
s (typically a simple Newtonian fluid) and the polymer/colloid contribution p . The exact
form of the stress tensor depends upon the details of the microstructural model. However, the
most common model for a polymer or colloid is a collection of discrete beads (with positions
ri ) for which the particle contribution to the bulk stress is given by the Kramers-Kirkwood
expression *N +
1 X
p = ri Fhi (16)
V i=1

where Fhi is the total hydrodynamic force exerted on bead i, V is the volume of the simulation
box and h i denotes an ensemble average. Equation (16) is a quite general result and is
still applicable for systems with constraints (e.g. bead-rod chains) and when hydrodynamic
interactions are taken into account (Bird et al., 1987; Morse, 2004).
While it is important to establish Equation (16) as a rather general form for the stress
tensor, it useful to consider the specific form of the tensor for restricted classes of models
and other commonly encountered cases. Consider first the canonical model employed in
Brownian dynamics for a polymer: a series of N beads connected by springs. The polymer
contribution to the stress is then
* N +
X
p = np Fsi + Fother


i Ri + np NkB T (17)
i=1

7
where Ri is the position of a bead relative to the center of mass of the chain, np is the
number density of polymer molecules, Fsi the net force by springs on bead i and Fother i
is the sum of all the non-hydrodynamic and non-spring forces exerted on bead i, such as
bead-bead excluded volume interactions or bending forces. Equation (17) is valid both when
hydrodynamic interactions are accounted for or neglected. The last term in Equation (17)
is the familiar ideal gas-like internal energy contribution from each bead which is a direct
result of applying the usual equilibration in momentum space and assuming an isotropic
friction tensor for a bead (Bird et al., 1987).
There can be computational advantages to using certain formulas over others to calculate
the stress tensor in particular simulations. To gain some appreciation of this consider the
steady-state rheology of a bead-rod (or bead-spring) chain in a simple shear flow u
x (r) = r
y
and neglect hydrodynamic interactions between beads. Using the Giesekus stress formula
the polymer contribution to the shear viscosity ( p = xy
p
/)
can be written
* N
+ * N +
np d X np X
p = Rxi Ryi + Ryi Ryi . (18)
2 dt i=1
2 i=1

However, because in this example we are only interested in the stress of the system at
steady-state we can immediately set the first term on the right side of Equation (18) to
zero and eliminate it from our simulation code. Not only is this a great simplification, but
now one has a simple expression which gives physical insight into the connection between
polymer viscosity and molecular configurations. For this example, the effective width of the
chain in the shear-gradient direction is directly related to the contribution to the viscosity.
Furthermore, simplifying the formula for the stress tensor can aid in the development of
scaling relations which generalize the trends (Doyle et al., 1997).
An extensive discussion of the various forms of the stress tensor can be found in Bird et al.
(1987). We note that extra care must be taken when handling polymers containing constraint
forces. In this case, a brute force calculation of the stress using Equation (16) will contain
some contributions which fluctuate about zero but are quite large (order t1/2 ). Various
efficient algorithms have been developed specifically for this case which essentially filter out
these uninteresting noise terms (Doyle et al., 1997; Morse, 2004).

Polymer Solutions in Hydrodynamic Flows

The ability to predict the dynamics and rheology of dilute polymer solutions in simple
linear flows is an important benchmark by which researchers evaluate BD simulations and
mesoscopic polymer models. In recent years it has become possible to directly observe single
polymer molecule dynamics in flow using double stranded DNA (Chu, 2003). DNA is a

8
 1.0

Mean Fractional Extension

Free extensional flow
0.8

Tethered shear flow

0.6

0.4
Free shear flow

0.2

0.0
0 20 40 60 80 100
Wi

Figure 2: Comparison of Brownian dynamics simulations (lines) to single molecule DNA

experiments (symbols). Mean fractional extension (stretch of the DNA scaled by the contour
length of the molecule) versus Weissenberg number (W i) is shown for tethered and free DNA
in linear hydrodynamic flows. The W i is the product of the shear rate (or elongation rate)
and the longest relaxation time of the DNA. Figure adapted from Doyle et al. (2000).

unique polymer in that its force-extension relation (entropic spring-like force) has been well
characterized and for large DNA is accurately described by the Marko-Siggia force law. To
date most experiments have been performed with a particular commercially available DNA
molecule, -DNA, which when stained with typical fluorescent dyes contains approximately
400 persistence lengths. Figure 2 shows a comparison between BD simulations using a free-
draining (no HI) bead-spring chain (spring law given by Equation (12)) for free and tethered
-DNA in simple linear flows. Excellent quantitative agreement between free-draining BD
simulations and experiments of -DNA in shear, elongational and mixed flows has been
attained (see for example Larson et al. (1999) and Doyle et al. (2000)). The ability to
neglect HI in the BD simulations and obtain quantitative agreement with experiments is
because -DNA contains a modest number of persistence lengths and so even when HI is
included there is little difference in the drag on a coiled and fully extended chain. Recent
experiments using extremely long DNA (genomic length, 20, 000 persistence lengths) have
demonstrated conformational hysteresis in extensional flows (Schroeder et al., 2003). The
authors show the qualitative nature of this behavior is captured only if HI is included in BD
simulations.
As it stands now, quantitative description of most trends in the stretching of DNA with

9
lengths up to 1000 persistence lengths can be captured without including HI in a sim-
ulation. Further simulations need to be done to quantitatively test HI when comparing
to larger genomic length DNA. Furthermore, when polymers are placed near hard surfaces
the hydrodynamic interaction between segments will change due to satisfying the no-slip
boundary condition at the surface. These wall effects can be taken into account by numeri-
cally solving for the Greens function for creeping flow in an arbitrary geometry (Jendrejack
et al., 2004). These modified hydrodynamic interactions are needed in a BD simulation
to predict the correct trend of shear induced migration in channels. However, it has been
shown that quantitative agreement between free-draining BD simulations and experiments
for the stretch and fluctuations of tethered -DNA is obtained if the drag coefficient is merely
adjusted (Doyle et al., 2000).
While the ability to observe single polymer molecules is very powerful, a measure of the
polymer contribution to the stress is not attained in such experiments. Recently it has
become possible to accurately measure transient extensional stresses of high molecular weight
polymer solutions using filament stretching rheometers (McKinley and Sridhar, 2002). The
stresses developed in these flows are challenging to model because the elongational nature of
the flow leads to large deformation of the initially coiled polymer and brings with this a large
contribution by the polymer to solution viscosity. Hsieh and Larson have performed the most
detailed study of the role in HI in BD simulations of polystyrene and DNA in extensional
flows (Hsieh et al., 2003). A method was developed to determine the HI parameters in the
model which on one hand will keep the number of springs to a modest number while on
the other hand will match the relaxation time or diffusivity of the experimental polymer
system and the drag on the fully stretch polymer (estimated from Batchelors formula).
They confirmed that including HI has little effect for -DNA. Hsieh and Larson find that
it is necessary to include HI in order to quantitatively match stress-strain behavior up to
strains of 6 in filament stretching experiments of polystyrene with a molecular weight of 2
million ( 5400 persistence lengths). However, the simulations do not properly predict the
experimental values for the high strain plateau in the stress.

Outlook

Brownian dynamics is a powerful technique to simulate nonequilibrium dynamics of polymers

and other soft matter. Efficient and stable algorithms have been developed which allow for
the simulation of a wide class of polymer models ranging from flexible bead-spring chains to
semiflexible bead-rod filaments. Furthermore, the strengths and deficiencies of springs repre-
senting a small number of persistence lengths is now well understood, at least for unconfined
polymers. Quantitative comparisons of BD simulations to the rheology and dynamics of
dilute polymer solutions show that our understanding of the importance and correct imple-
mentation of hydrodynamic interactions into a simulation is continuing to evolve.

10
With the growing importance of processing biological and other complex fluids in micro
and even nanochannels, the BD simulation technique will continue to advance to properly
treat molecules in tight spaces. Further quantitative comparisons to single molecule DNA
experiments, both in ideal bulk flows and in microfluidic devices, will be critical in helping
us to evaluate the state of the art in Brownian dynamics simulations.

Bibliography
Bird, R. B., Curtiss, C. F., Armstrong, R. C. and Hassager, O., 1987. Dynamics of Polymeric
Liquids, Volume 2 . 2nd edn. John Wiley and Sons, New York.

Chu, S., 2003. Biology and polymer physics at the single molecule level. Phil. Trans. R. Soc.
Lond. A 361, 689.

Doyle, P. S., Gast, A. P. and Shaqfeh, E. S. G., 1997. Dynamic simulation of freely draining
flexible polymers in steady linear flows. J. Fluid Mech. 334, 251.

Doyle, P. S., Ladoux, B. and Viovy, J. L., 2000. Dynamics of a tethered polymer in shear
flow. Phys. Rev. Lett. 84, 4769.

Fixman, M., 1986. Construction of Langevin forces in the simulation of hydrodynamic

interaction. Macromolecules 19, 1204.

Grassia, P. S., Hinch, E. J. and Nitsche, L. C., 1995. Computer simulations of Brownian
motion of complex systems. J. Fluid Mech. 282, 373.

Hsieh, C.-C., Li, L. and Larson, R. G., 2003. Modeling hydrodynamic interaction in Brownian
dynamics: simulations of extensional flows of dilute solutions of DNA and polystyrene. J.
non-Newtonian Fluid Mech. 113, 147.

Jendrejack, R. M., Schwartz, D. C., de Pablo, J. J. and Graham, M. D., 2004. Shear-induced
migration in flowing polymer solutions: simulation of long-chain DNA in microchannels.
J. Chem. Phys. 120, 2513.

Larson, R. G., Hu, H., Smith, D. E. and Chu, S., 1999. Brownian dynamics simulations of
DNA molecules in an extensional flow field. J. Rheol. 43, 267.

Marko, E. and Siggia, E. D., 1995. Stretching DNA. Macromolecules 28, 8759.

McKinley, G. H. and Sridhar, T., 2002. Filament stretching rheometry of complex fluids.
Ann. Rev. Fluid Mech. 34, 375.

11
Montesi, A., Pasquali, M. and MacKintosh, F. C., 2004. Collapse of a semiflexible polymer
in poor solvent. Phys. Rev. E 69, 021916.

Morse, D. C., 2004. Theory of constrained Brownian motion. Adv. Chem. Physics 128, 65.

Ottinger, H. C., 1996. Stochastic processes in polymeric fluids. Springer-Verlag, New York.

Schroeder, C. M., Babcock, H. P., Shaqfeh, E. S. G. and Chu, S., 2003. Observation of
polymer hysteresis in extensional flow. Science 3001, 1515.

Underhill, P. T. and Doyle, P. S., 2004. On the coarse-graining of polymers into bead-spring
chains. J. non-Newtonian Fluid Mech. in press.

Yamakawa, H., 1997. Helical Wormlike Chains in Polymer Solutions. Springer, Berlin.

Patrick S. Doyle and Patrick T. Underhill

Department of Chemical Engineering
Massachusetts Institute of Technology
Cambridge, Massachusetts, USA
pdoyle@mit.edu; ptu@mit.edu

12
Figure 1: Schematic of the canonical polymer model used in Brownian dynamics simulations:
a bead-spring chain. The continuous curve is the real polymer that the bead-spring chain
is meant to represent.

13
 1.0
Mean Fractional Extension

Free extensional flow

0.8

Tethered shear flow

0.6

0.4
Free shear flow

0.2

0.0
0 20 40 60 80 100
Wi

Figure 2: Comparison of Brownian dynamics simulations (lines) to single molecule DNA

experiments (symbols). Mean fractional extension (stretch of the DNA scaled by the contour
length of the molecule) versus Weissenberg number (W i) is shown for tethered and free DNA
in linear hydrodynamic flows. The W i is the product of the shear rate (or elongation rate)
and the longest relaxation time of the DNA. Figure adapted from Doyle et al. (2000).

14